#THINKFUTURE
Series: Building the Future on a Cracked Foundation
About this series
Artificial intelligence has become one of the defining leadership topics of our time. Organisations are investing billions in technologies that promise greater productivity, lower costs and faster decision-making. Those investments matter, but this series begins with a different question:
What if the greatest challenge of AI is not how it changes work, but how it changes the way organisations develop capable people?
In this THINK FUTURE series, Building the Future on a Cracked Foundation, we explore that question through the lenses of research, organisational psychology and practical leadership. Each essay challenges a common assumption and invites leaders to think differently about capability, work and organisational success.
The task that appeared to have no value
Imagine a professional services firm reviewing the work performed by its junior employees.
The review team identifies several activities that appear inefficient. Junior analysts spend hours gathering information, checking calculations, preparing first drafts and revising work after feedback from more experienced colleagues. Artificial intelligence can now complete much of this work more quickly and, in some cases, to a higher initial standard.
The conclusion seems obvious. If AI can produce the same output in minutes, why should a junior employee spend hours producing it?
The firm automates much of the work. Productivity improves, turnaround times decline and senior employees spend less time correcting basic errors. From an operational perspective, the decision appears successful.
Several years later, however, a different problem emerges. The firm has capable senior professionals and efficient AI systems, but too few people are ready to move into the roles between them. Employees who were protected from routine work can produce polished outputs with AI assistance, yet some struggle to explain the reasoning behind those outputs, recognise unusual patterns or know when the technology is wrong.
The organisation removed work that appeared to have little productive value. Only later did it discover that the same work had carried considerable developmental value.
This is a hypothetical scenario, but the question it raises is becoming increasingly real:
What if leaders have misunderstood what work produces?
The output leaders can see
Most organisations treat work as a production system. People apply their knowledge and effort to create something the organisation values. They advise a client, resolve a customer complaint, prepare a financial report, repair equipment, design a product or make a decision.
This understanding is not wrong. Organisations must produce useful outcomes, and productivity affects their ability to compete, invest and survive. Leaders therefore measure what work delivers: output, quality, cost, speed, revenue, customer satisfaction and return on investment.
What these measures seldom reveal is what happens to the person performing the work.
A junior accountant preparing a reconciliation is not only completing a financial task. That person is learning how transactions flow through a business, which discrepancies matter and how apparently insignificant anomalies can signal a larger problem. A graduate engineer performing calculations is not only producing numbers. The engineer is developing an intuitive understanding of relationships, constraints and the conditions under which an answer should not be trusted.
The visible output belongs to the organisation. The less visible output remains within the person as growing knowledge, confidence, pattern recognition and judgement.
Work therefore does more than convert existing capability into present performance. Under the right conditions, it also develops the capability on which future performance will depend.
Work has always been a learning environment
Much of professional development does not happen during formal training. It occurs while people solve unfamiliar problems, observe experienced colleagues, receive feedback, correct mistakes and reflect on what happened.
A major review of workplace learning describes learning at work as encompassing incidental and informal learning, intentional but non-formal development, and formal training. This matters because knowledge and professional competence are not acquired through a single mechanism. They emerge through the interaction between what people know, what they do and the social environments in which they work. Tynjälä’s review of workplace learning established this integrative view well before generative AI entered organisational life.
More recent research continues to identify experience, experimentation, observation, feedback and reflection as central components of informal workplace learning. This learning is often difficult to see precisely because it is embedded in the work itself. Employees may be developing capability without thinking of the activity as training. Research by Decius and colleagues describes informal learning as occurring through reflection, feedback and the completion of work tasks, while a recent systematic review found that workplace learning depends on a combination of individual and contextual conditions rather than exposure to experience alone. Wijga and colleagues’ systematic review reinforces the importance of the environment in which the work takes place.
This distinction is important. Experience does not automatically produce expertise. Repeating the same task without challenge, feedback or reflection may reinforce poor habits rather than improve performance. Work becomes developmental when it stretches a person appropriately, makes the consequences of decisions visible, provides credible feedback and creates opportunities to adjust future action.
A 2025 realist review of adaptive expertise development identified five relevant features of work-based learning: challenging and integrated contexts, reflective practice, interaction with others, guidance, and characteristics of the learner. The review concluded that work-based learning can help people develop the ability to respond to unfamiliar circumstances when it requires them to integrate knowledge, adapt established approaches and develop new ones. Groenier and colleagues’ realist review therefore supports a more precise conclusion than the simple claim that people learn by doing.
People develop capability through the right kind of doing.
Why apparently routine work matters
The developmental value of work is not always proportionate to its visible complexity.
Some early-career tasks are repetitive and inefficient. They may be frustrating for the employee and costly for the organisation. Automating them can be entirely appropriate. The mistake is to assume that a task has no developmental value simply because it has limited productive value.
Routine work can expose people to the normal patterns of a profession. Repetition helps them see what usually happens, which is often necessary before they can recognise what is unusual. First drafts reveal how someone structures a problem before a manager improves the answer. Preliminary calculations force an engineer to engage with the relationships that software can otherwise conceal. Initial customer interactions help an employee learn how similar problems can require different responses.
The correction process matters as much as the first attempt. When a junior employee produces work, receives feedback and revises it, the difference between the initial and improved versions becomes a source of learning. If the first attempt is removed entirely, the organisation may also remove the opportunity to observe how that person thinks.
This does not mean that organisations should preserve inefficient work for its own sake. Nor does it mean that every employee must manually perform every task previously completed by earlier generations. It means that leaders should understand what people learn through a task before deciding that its only value lies in the output it produces.
The relevant question is not simply, “Can this task be automated?”
It is also, “What was this task teaching?”
AI exposes a weakness in the way we measure work
Generative AI makes this issue more urgent because it can separate the quality of an output from the capability of the person presenting it.
In a large field study involving customer-support agents, access to a generative AI assistant improved productivity, with the largest gains occurring among less experienced and lower-skilled employees. The researchers found suggestive evidence that the system helped disseminate the practices of more capable workers and accelerated movement along the experience curve. This is an important demonstration of AI’s developmental potential. Properly designed systems may give employees access to guidance that was previously inconsistent or unavailable. Brynjolfsson, Li and Raymond’s study was published in The Quarterly Journal of Economics in 2025.
The finding should not, however, be interpreted as proof that every AI-assisted improvement represents durable human capability. There is a difference between performing better with support and becoming more capable of performing without it.
Evidence from education illustrates why that distinction matters. In a large field experiment in mathematics, students given access to a general-purpose generative AI interface performed better during supported practice but subsequently performed worse when the assistance was removed. A more carefully designed AI tutor, with safeguards intended to support learning rather than simply supply answers, largely avoided that negative effect. Bastani and colleagues’ 2025 PNAS study was conducted in education rather than the workplace, so its findings should not be transferred uncritically to employees. It nevertheless demonstrates an important principle: an intervention can improve assisted performance while weakening independent learning if it removes too much of the thinking the learner needs to practise.
AI can therefore support capability development or bypass it. The result depends on how the technology, task and learning process are designed.
If an employee uses AI to compare approaches, test assumptions, receive feedback and examine the reasoning behind a recommendation, the technology may enrich the learning experience. If the employee merely requests an answer, accepts it and passes it forward, the output may improve without a comparable improvement in understanding.
This is why measuring AI adoption primarily through speed and output creates an incomplete picture. Those measures reveal what the human and the technology produced together today. They do not necessarily reveal what the human will be capable of doing, questioning or judging tomorrow.
From production system to capability-development system
Leaders do not need to choose between productivity and development. They need to stop treating the two as if they are unrelated.
An organisation that views work only as a production system asks how tasks can be completed faster, more consistently and at lower cost. An organisation that also recognises work as a capability-development system asks what knowledge and judgement people acquire while those tasks are performed.
This broader view does not require leaders to retain obsolete processes. It requires them to identify the learning embedded in existing work and redesign that learning intentionally when the work changes.
Before automating or substantially redesigning a task, leaders can examine four questions:
- What does the task produce? Identify its immediate operational value, including output, quality, cost and customer impact.
- What does the task teach? Determine whether it develops foundational knowledge, pattern recognition, judgement, professional identity, relationships or an understanding of organisational context.
- How does the learning occur? Identify the role of practice, observation, feedback, reflection, social interaction and progressively greater responsibility.
- How will that learning be preserved or improved? Decide whether AI will support the original learning process or whether new experiences, simulations, mentoring arrangements and accountability will be needed.
These questions should not become an excuse for preserving poor work design. Some tasks are neither productive nor developmental. Others may teach lessons that can be learned more effectively through AI-supported practice, simulation or structured exposure to more complex problems.
The purpose is not to protect tasks. It is to protect and improve the development of capability.
The productivity paradox leaders may create
An organisation can become more productive at the level of individual tasks while becoming less capable at the level of its future workforce.
The contradiction may remain hidden for years. AI can help inexperienced employees produce work that resembles the output of experienced professionals. Senior employees may then conclude that fewer junior people are required. Recruitment declines, entry-level work contracts and the organisation becomes increasingly dependent on a smaller group of experts.
The immediate economics may look attractive. What the calculation may omit is that experienced professionals do not appear fully formed in the labour market. They develop through successive cycles of practice, feedback, increasing responsibility and exposure to consequences.
If organisations reduce entry-level opportunities without creating credible alternative pathways, they may weaken the pipeline into the very roles they will later struggle to fill. The risk is not simply a shortage of people at the bottom of the organisational hierarchy. It is a shortage of sufficiently developed people in the middle, where organisations often rely on experienced professionals to interpret complexity, supervise others, manage exceptions and translate strategic intent into action.
This is the deeper crack in the foundation. An organisation can automate the work through which people once developed, assume that formal training will fill the gap, and discover too late that knowledge about a profession is not equivalent to judgement within it.
A different leadership conversation
The first essay in this series asked leaders to consider what capability their organisations might stop developing when AI changes work. This second essay takes the argument further.
Work should not be protected simply because it existed in the past. Neither should it be removed simply because a machine can produce its visible output more efficiently. Leaders must first understand the full value of the work, including the capability it develops in the person performing it.
That changes the objective of AI-enabled work design. The goal is no longer merely to allocate each task to the fastest or cheapest performer. It is to create a system in which human and technological capability reinforce one another over time.
Some work should be automated. Some should be augmented. Some should remain a space in which people think, practise and learn. Other work should be redesigned so that AI creates better feedback, richer experimentation and earlier exposure to complex problems.
The correct choice cannot be made by examining productivity alone.
THINK FUTURE Principle
Work does not only produce results through capable people. It also helps produce the capable people needed to deliver future results.
The leadership challenge is therefore not to preserve yesterday’s tasks. It is to ensure that tomorrow’s work continues to develop the knowledge, judgement and adaptive capability the organisation will need.
As AI makes more work easier to complete, leaders should ask:
Are we improving how quickly people produce the answer, or are we also improving how they become capable of judging whether the answer is right?
Coming next
Are We Mistaking Information for Expertise?
Artificial intelligence can give employees immediate access to information, analysis and expert-looking answers. It can help less experienced people produce work that might previously have required years of professional development.
But access to an answer is not the same as understanding it. Producing convincing work is not the same as knowing whether it can be trusted.
In the next essay, we examine a distinction that may become increasingly difficult for leaders to see:
When AI improves what people can produce, how will organisations know whether human expertise is developing too?